Latest post
Opus 5 is a great employee and a frustrating boss
Opus 5 matches Fable 5 on the benchmarks at half the token price, yet the first week of complaints is real too. Where it shines, and where it needs a boss.
Hands-on notes on AI and software development
Model releases, benchmarks, experiments and guides for developers who work with AI every day. Written from real use, with sources for every claim.
GPT-5.6 Sol finds everything. That's the problem
GPT-5.6 Sol found 400 audit issues where Fable 5 found 70, then over-engineered the fixes. Same trait, both results. The working contract that tames it.